How it works — Agentkin

41|41|

Week 0 — Discovery (20 minutes)

42|42|

You share your screen. We watch you work for fifteen minutes — emails, follow-ups, internal hand-offs, where you copy-paste data between tools, where you wait on someone to do a thing manually.

43|43|

At the end of the call we propose three high-leverage workflows we'd build first. You pick the one (or two) that resonate. We agree on a start date, sign a one-page SaaS contract, you go.

44|44| 45|45|

Day 1-2 — The Build (48 hours)

46|46|

We provision an isolated cloud computer in a Netherlands datacenter. Your agent lives there — not on a shared platform, not behind someone else's API. We name it (you pick), give it its own email address, and connect it to your tools (Gmail, Slack, Notion, your CRM, your billing system — anything with an API).

47|47|

Then we teach it the chosen workflows. Not by writing rigid scripts — by giving it the same context a new hire would get: examples of past work, your client list, your house style, what good looks like.

48|48|

By hour 48 you have:

49|49|

A Telegram chat (or our web dashboard, your choice) where you talk to your agent
The first workflow live and producing output
A Loom from us walking you through how to use it

54|54| 55|55|

Week 1-onward — Continuous tuning

56|56|

This is where most AI projects die. Yours won't.

57|57|

You have a shared dashboard with four columns: Backlog · To-Do · Doing · Done. Whenever you think of something — "I wish it could also do X", "this part isn't right", "add this skill" — you drop a card in the Backlog. We promise 1-2 changes shipped every 48 hours.

58|58|

Every week, you get a Friday Loom from us: what shipped this week, what we noticed about how you use the agent, what we're thinking about next.

59|59| 60|60|

What we handle, what you handle

61|61| 62|62| 63|63| 64|64| 65|65| 66|66| 67|67| 68|68| 69|69| 70|70| 71|71| 72|72| 73|73| 74|74|

	You	Agentkin
Telling us what you need	✓
Sharing access to tools	✓
Reviewing agent output occasionally	✓
Picking the model (GPT, Claude, etc.)		✓
Building the agent		✓
Token costs		✓
Hosting + infrastructure		✓
Monitoring 24/7 + auto-restart on failure		✓
Fixing things that break		✓
Weekly tuning + new skills		✓
Daily security updates + backups		✓

75|75| 76|76|

The stack (so you know it's serious)

77|77|

Your agent runs on its own dedicated cloud VPS — not shared with other customers, not on a public AI platform. Stack:

78|78|

Agent runtime — Hermes by Nous Research (open source, self-improving)
Model — Claude Sonnet 4.6 by default; we'll swap to whatever's best in 6 months
Tool layer — Composio (one connector → 1000+ apps, secure auth)
Memory — Per-customer Obsidian vault (your agent never forgets)
Monitoring — Internal watchdog + alerts. We get paged before you do.

85|85|

You don't need to remember any of this. It's our problem. But if you ever want to take everything and walk, you can — your agent + its memory + its skills are yours. We hand you the keys.

86|86| 87|87|

88|88|

Ready to see what your agent looks like?

89|89|

20-minute call. We screen-share, you show us where you bleed time.

90|90| Book your discovery call 91|91|

92|92|